Now that the dust has settled, whose polls were the best predictors of the actual midterm results? Nate Silver ranks them by accuracy and party bias over the last 21 days of the election cycle. The most accurate, according to his calculations: Quinnipiac, which “showed little bias,” he writes in the New York Times. The worst: Rasmussen, which scored poorly on both accuracy and bias.
Silver analyzed the margin between the two top candidates’ percentage of the vote: If a poll predicted a Dem would win by 2 percentage points and he lost by 3, Silver assigned it a 5-point error. Rasmussen “overestimated the standing of the Republican candidate by almost 4 points on average,” he notes, while Quinnipiac overestimated Republican success by just 0.7 points and “missed the final margin between the candidates” by an average of just 3.3 points. Rasmussen's final average was off 5.8 points, and a whopping 40 points in the case of the Hawaii Senate race. SurveyUSA was the runner-up for accuracy, with an average error of 3.5 points.